X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=acf36f5c05f5317cf8d9d5a629108d98afe0f77e;hb=f37bce8a573dfc5aac1b9f51f4d5c8314ba05d30;hp=554a26da040a6f7646b7150b30caab7f2b2e7939;hpb=33f95f7ae7300f93433bc781845594d292a18b67;p=fs%2Flustre-release.git

diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c
index 554a26d..acf36f5 100644
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -285,6 +285,76 @@ osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 	return idc;
 }
 
+static void osd_idc_dump_lma(const struct lu_env *env,
+				struct osd_device *osd,
+				unsigned long ino,
+				bool check_in_oi)
+{
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
+	const struct lu_fid *fid;
+	struct osd_inode_id lid;
+	struct inode *inode;
+	int rc;
+
+	inode = osd_ldiskfs_iget(osd_sb(osd), ino);
+	if (IS_ERR(inode)) {
+		CERROR("%s: can't get inode %lu: rc = %d\n",
+		       osd->od_svname, ino, (int)PTR_ERR(inode));
+		return;
+	}
+	if (is_bad_inode(inode)) {
+		CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
+		goto put;
+	}
+	rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
+	if (rc) {
+		CERROR("%s: can't get LMA for %lu: rc = %d\n",
+		       osd->od_svname, ino, rc);
+		goto put;
+	}
+	fid = &loa->loa_lma.lma_self_fid;
+	LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
+		      PFID(fid), ino, (unsigned)inode->i_generation);
+	if (!check_in_oi)
+		goto put;
+	rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
+	if (rc) {
+		CERROR("%s: can't lookup "DFID": rc = %d\n",
+		       osd->od_svname, PFID(fid), rc);
+		goto put;
+	}
+	LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
+		      PFID(fid), lid.oii_ino, lid.oii_gen);
+put:
+	iput(inode);
+}
+
+static void osd_idc_dump_debug(const struct lu_env *env,
+				struct osd_device *osd,
+				const struct lu_fid *fid,
+				unsigned long ino1,
+				unsigned long ino2)
+{
+	struct osd_inode_id lid;
+
+	int rc;
+
+	rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
+	if (!rc) {
+		LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
+			osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
+		osd_idc_dump_lma(env, osd, lid.oii_ino, false);
+	} else {
+		CERROR("%s: can't lookup "DFID": rc = %d\n",
+		       osd->od_svname, PFID(fid), rc);
+	}
+	if (ino1)
+		osd_idc_dump_lma(env, osd, ino1, true);
+	if (ino2)
+		osd_idc_dump_lma(env, osd, ino2, true);
+}
+
 /*
  * lookup mapping for given FID and fill it from the given object.
  * the object is lolcal by definition.
@@ -302,7 +372,12 @@ static int osd_idc_find_and_init(const struct lu_env *env,
 		if (obj->oo_inode == NULL)
 			return 0;
 		if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
-			LASSERT(idc->oic_lid.oii_ino == 0);
+			if (idc->oic_lid.oii_ino) {
+				osd_idc_dump_debug(env, osd, fid,
+						   idc->oic_lid.oii_ino,
+						   obj->oo_inode->i_ino);
+				return -EINVAL;
+			}
 			idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 			idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 		}
@@ -392,12 +467,11 @@ int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 		lustre_loa_swab(loa, true);
 		/* Check LMA compatibility */
 		if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
-			CWARN("%s: unsupported incompat LMA feature(s) %#x "
-			      "for fid = "DFID", ino = %lu\n",
+			rc = -EOPNOTSUPP;
+			CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
 			      osd_ino2name(inode),
 			      lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
-			      PFID(&lma->lma_self_fid), inode->i_ino);
-			rc = -EOPNOTSUPP;
+			      PFID(&lma->lma_self_fid), inode->i_ino, rc);
 		}
 	} else if (rc == 0) {
 		rc = -ENODATA;
@@ -442,10 +516,11 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 		iput(inode);
 		inode = ERR_PTR(-ESTALE);
 	} else if (is_bad_inode(inode)) {
-		CWARN("%s: bad inode: ino = %u\n",
-		      osd_dev2name(dev), id->oii_ino);
+		rc = -ENOENT;
+		CWARN("%s: bad inode: ino = %u: rc = %d\n",
+		      osd_dev2name(dev), id->oii_ino, rc);
 		iput(inode);
-		inode = ERR_PTR(-ENOENT);
+		inode = ERR_PTR(rc);
 	} else if ((rc = osd_attach_jinode(inode))) {
 		iput(inode);
 		inode = ERR_PTR(rc);
@@ -484,7 +559,7 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 		if (!rc2) {
 			fid = &loa->loa_lma.lma_self_fid;
 		} else if (rc2 == -ENODATA) {
-			if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
+			if (unlikely(is_root_inode(parent))) {
 				fid = &info->oti_fid3;
 				lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 			} else if (!osd->od_is_ost && osd->od_index == 0) {
@@ -529,7 +604,7 @@ osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 	if (!rc) {
 		*fid = loa->loa_lma.lma_self_fid;
 	} else if (rc == -ENODATA) {
-		if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
+		if (unlikely(is_root_inode(inode)))
 			lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 		else
 			lu_igif_build(fid, inode->i_ino, inode->i_generation);
@@ -858,7 +933,6 @@ struct osd_check_lmv_buf {
 	struct dir_context ctx;
 	struct osd_thread_info *oclb_info;
 	struct osd_device *oclb_dev;
-	struct osd_idmap_cache *oclb_oic;
 	int oclb_items;
 	bool oclb_found;
 };
@@ -884,7 +958,6 @@ static int osd_stripe_dir_filldir(void *buf,
 	struct lu_fid *fid = &oti->oti_fid3;
 	struct osd_inode_id *id = &oti->oti_id3;
 	struct osd_device *dev = oclb->oclb_dev;
-	struct osd_idmap_cache *oic = oclb->oclb_oic;
 	struct inode *inode;
 
 	oclb->oclb_items++;
@@ -907,10 +980,7 @@ static int osd_stripe_dir_filldir(void *buf,
 
 	iput(inode);
 	osd_add_oi_cache(oti, dev, id, fid);
-	oic->oic_fid = *fid;
-	oic->oic_lid = *id;
-	oic->oic_dev = dev;
-	osd_oii_insert(dev, oic, true);
+	osd_oii_insert(dev, fid, id, true);
 	oclb->oclb_found = true;
 
 	return 1;
@@ -953,7 +1023,7 @@ static int osd_stripe_dir_filldir(void *buf,
  *    the correct OI mapping for the slave MDT-object.
  */
 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
-			 struct inode *inode, struct osd_idmap_cache *oic)
+			 struct inode *inode)
 {
 	struct lu_buf *buf = &oti->oti_big_buf;
 	struct dentry *dentry = &oti->oti_obj_dentry;
@@ -963,7 +1033,6 @@ static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 		.ctx.actor = osd_stripe_dir_filldir,
 		.oclb_info = oti,
 		.oclb_dev = dev,
-		.oclb_oic = oic,
 		.oclb_found = false,
 	};
 	int rc = 0;
@@ -1016,10 +1085,10 @@ again:
 
 out:
 	if (rc < 0)
-		CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
-		       DFID": rc = %d\n", osd_ino2name(inode),
-		       inode->i_ino, inode->i_generation,
-		       PFID(&oic->oic_fid), rc);
+		CDEBUG(D_LFSCK,
+		       "%s: cannot check LMV, ino = %lu/%u: rc = %d\n",
+		       osd_ino2name(inode), inode->i_ino, inode->i_generation,
+		       rc);
 	else
 		rc = 0;
 
@@ -1052,7 +1121,13 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 
 	LINVRNT(osd_invariant(obj));
 	LASSERT(obj->oo_inode == NULL);
-	LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
+
+	if (fid_is_sane(fid) == 0) {
+		CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
+		       PFID(fid));
+		dump_stack();
+		RETURN(-EINVAL);
+	}
 
 	dev = osd_dev(ldev);
 	scrub = &dev->od_scrub.os_scrub;
@@ -1171,16 +1246,15 @@ trigger:
 		if (scrub->os_partial_scan && !scrub->os_in_join)
 			goto join;
 
-		osd_add_oi_cache(info, dev, id, fid);
 		if (IS_ERR_OR_NULL(inode) || result) {
-			osd_oii_insert(dev, oic, result == -ENOENT);
+			osd_oii_insert(dev, fid, id, result == -ENOENT);
 			GOTO(out, result = -EINPROGRESS);
 		}
 
 		LASSERT(remote);
 		LASSERT(obj->oo_inode == inode);
 
-		osd_oii_insert(dev, oic, true);
+		osd_oii_insert(dev, fid, id, true);
 		goto found;
 	}
 
@@ -1202,16 +1276,15 @@ join:
 	if (rc1 && rc1 != -EALREADY)
 		GOTO(out, result = -EREMCHG);
 
-	osd_add_oi_cache(info, dev, id, fid);
 	if (IS_ERR_OR_NULL(inode) || result) {
-		osd_oii_insert(dev, oic, result == -ENOENT);
+		osd_oii_insert(dev, fid, id, result == -ENOENT);
 		GOTO(out, result = -EINPROGRESS);
 	}
 
 	LASSERT(remote);
 	LASSERT(obj->oo_inode == inode);
 
-	osd_oii_insert(dev, oic, true);
+	osd_oii_insert(dev, fid, id, true);
 	goto found;
 
 check_lma:
@@ -1308,6 +1381,8 @@ check_lma:
 
 	if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
 		result = -EREMCHG;
+		osd_scrub_refresh_mapping(info, dev, fid, id, DTO_INDEX_DELETE,
+					  true, 0, NULL);
 		goto trigger;
 	}
 
@@ -1342,7 +1417,7 @@ found:
 
 	if (S_ISDIR(inode->i_mode) &&
 	    (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
-		osd_check_lmv(info, dev, inode, oic);
+		osd_check_lmv(info, dev, inode);
 
 	result = osd_attach_jinode(inode);
 	if (result)
@@ -1757,6 +1832,7 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
 	th->th_dev = d;
 	th->th_result = 0;
 	oh->ot_credits = 0;
+	oh->oh_declared_ext = 0;
 	INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
 	INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
 	INIT_LIST_HEAD(&oh->ot_trunc_locks);
@@ -2040,6 +2116,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
 	if (unlikely(remove_agents != 0))
 		osd_process_scheduled_agent_removals(env, osd);
 
+	LASSERT(oti->oti_ins_cache_depth > 0);
 	oti->oti_ins_cache_depth--;
 	/* reset OI cache for safety */
 	if (oti->oti_ins_cache_depth == 0)
@@ -2346,13 +2423,12 @@ static void osd_conf_get(const struct lu_env *env,
 					OBD_CKSUM_T10IP512 :
 					OBD_CKSUM_T10IP4K;
 			} else {
-				CERROR("%s: unsupported checksum type of "
-				       "T10PI type '%s'",
+				CERROR("%s: unsupported checksum type of T10PI type '%s'\n",
 				       d->od_svname, name);
 			}
 
 		} else {
-			CERROR("%s: unsupported T10PI type '%s'",
+			CERROR("%s: unsupported T10PI type '%s'\n",
 			       d->od_svname, name);
 		}
 	}
@@ -2808,9 +2884,9 @@ static int osd_declare_attr_set(const struct lu_env *env,
 		bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
 
 		if (!ignore_edquot)
-			CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
-			       "(the quota space is %lld)\n",
-			       obj->oo_inode->i_sb->s_id, attr->la_uid,
+			CDEBUG(D_QUOTA,
+			       "%s: enforce quota on UID %u, GID %u (quota space is %lld)\n",
+			       osd_ino2name(obj->oo_inode), attr->la_uid,
 			       attr->la_gid, bspace);
 
 		/* USERQUOTA */
@@ -3003,9 +3079,8 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
 
 		rc = dquot_transfer(inode, &iattr);
 		if (rc) {
-			CERROR("%s: quota transfer failed: rc = %d. Is quota "
-			       "enforcement enabled on the ldiskfs "
-			       "filesystem?\n", inode->i_sb->s_id, rc);
+			CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
+			       osd_ino2name(inode), rc);
 			return rc;
 		}
 	}
@@ -3019,9 +3094,8 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
 		rc = -ENOTSUPP;
 #endif
 		if (rc) {
-			CERROR("%s: quota transfer failed: rc = %d. Is project "
-			       "enforcement enabled on the ldiskfs "
-			       "filesystem?\n", inode->i_sb->s_id, rc);
+			CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
+			       osd_ino2name(inode), rc);
 			return rc;
 		}
 	}
@@ -3145,12 +3219,21 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
 	struct osd_thandle *oth;
 	struct dt_object *parent = NULL;
 	struct inode *inode;
-	uid_t owner[2] = {0, 0};
+	struct iattr iattr = {
+		.ia_valid = ATTR_UID | ATTR_GID |
+			    ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
+		.ia_ctime.tv_sec = attr->la_ctime,
+		.ia_mtime.tv_sec = attr->la_mtime,
+		.ia_atime.tv_sec = attr->la_atime,
+		.ia_uid = GLOBAL_ROOT_UID,
+		.ia_gid = GLOBAL_ROOT_GID,
+	};
+	const struct osd_timespec omit = { .tv_nsec = UTIME_OMIT };
 
 	if (attr->la_valid & LA_UID)
-		owner[0] = attr->la_uid;
+		iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
 	if (attr->la_valid & LA_GID)
-		owner[1] = attr->la_gid;
+		iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
 
 	LINVRNT(osd_invariant(obj));
 	LASSERT(obj->oo_inode == NULL);
@@ -3170,10 +3253,18 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
 	    !dt_object_remote(hint->dah_parent))
 		parent = hint->dah_parent;
 
+	/* if a time component is not valid set it to UTIME_OMIT */
+	if (!(attr->la_valid & LA_CTIME))
+		iattr.ia_ctime = omit;
+	if (!(attr->la_valid & LA_MTIME))
+		iattr.ia_mtime = omit;
+	if (!(attr->la_valid & LA_ATIME))
+		iattr.ia_atime = omit;
+
 	inode = ldiskfs_create_inode(oth->ot_handle,
 				     parent ? osd_dt_obj(parent)->oo_inode :
 					      osd_sb(osd)->s_root->d_inode,
-				     mode, owner);
+				     mode, &iattr);
 	if (!IS_ERR(inode)) {
 		/* Do not update file c/mtime in ldiskfs. */
 		inode->i_flags |= S_NOCMTIME;
@@ -3460,6 +3551,9 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
 
 	LASSERT(obj->oo_inode != NULL);
 
+	if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
+		return -ENOSPC;
+
 	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle);
 	osd_trans_exec_op(env, th, OSD_OT_INSERT);
@@ -3818,6 +3912,9 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
 	__u32 saved_nlink = dir->i_nlink;
 	int rc;
 
+	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
+		return -ENOSPC;
+
 	dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
 	osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
 
@@ -3858,7 +3955,15 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	struct osd_thread_info *info = osd_oti_get(env);
 	struct inode *local;
 	struct osd_thandle *oh;
-	uid_t own[2] = {0, 0};
+	struct iattr iattr = {
+		.ia_valid = ATTR_UID | ATTR_GID |
+			    ATTR_CTIME | ATTR_MTIME | ATTR_ATIME,
+		.ia_ctime.tv_nsec = UTIME_OMIT,
+		.ia_mtime.tv_nsec = UTIME_OMIT,
+		.ia_atime.tv_nsec = UTIME_OMIT,
+		.ia_uid = GLOBAL_ROOT_UID,
+		.ia_gid = GLOBAL_ROOT_GID,
+	};
 	int rc;
 
 	ENTRY;
@@ -3867,7 +3972,8 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle->h_transaction != NULL);
 
-	local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type, own);
+	local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode,
+				     type, &iattr);
 	if (IS_ERR(local)) {
 		CERROR("%s: create local error %d\n", osd_name(osd),
 		       (int)PTR_ERR(local));
@@ -3887,8 +3993,8 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	 */
 	if (S_ISLNK(type)) {
 		BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
-		rc = snprintf((char *)LDISKFS_I(local)->i_data,
-			      LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
+		rc = scnprintf((char *)LDISKFS_I(local)->i_data,
+			       LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
 
 		i_size_write(local, rc);
 		LDISKFS_I(local)->i_disksize = rc;
@@ -3901,9 +4007,8 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	    i_projid_read(pobj->oo_inode) != 0) {
 		rc = osd_transfer_project(local, 0, th);
 		if (rc) {
-			CERROR("%s: quota transfer failed: rc = %d. Is project "
-			       "quota enforcement enabled on the ldiskfs "
-			       "filesystem?\n", local->i_sb->s_id, rc);
+			CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
+			       osd_ino2name(local), rc);
 			RETURN(ERR_PTR(rc));
 		}
 	}
@@ -4067,8 +4172,21 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt,
 			obj->oo_dt.do_body_ops = &osd_body_ops;
 	}
 
-	if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
+	if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
+		struct inode *inode = obj->oo_inode;
+
 		result = __osd_oi_insert(env, obj, fid, th);
+		if (result && inode) {
+			spin_lock(&obj->oo_guard);
+			clear_nlink(inode);
+			spin_unlock(&obj->oo_guard);
+			osd_dirty_inode(inode, I_DIRTY_DATASYNC);
+			ldiskfs_set_inode_state(inode,
+						LDISKFS_STATE_LUSTRE_DESTROY);
+			iput(inode);
+			obj->oo_inode = NULL;
+		}
+	}
 
 	/*
 	 * a small optimization - dt_insert() isn't usually applied
@@ -4092,6 +4210,7 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
 			       struct thandle *handle)
 {
 	struct osd_thandle *oh;
+	int rc;
 
 	/* it's possible that object doesn't exist yet */
 	LASSERT(handle != NULL);
@@ -4102,9 +4221,10 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
 	osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
 			     osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
 
-	osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
+	rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
+				   osd_dt_obj(dt));
 
-	return 0;
+	return rc;
 }
 
 /*
@@ -4834,6 +4954,10 @@ static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
 	return 0;
 }
 
+static bool osd_check_stale(struct dt_object *dt)
+{
+	return false;
+}
 /*
  * Index operations.
  */
@@ -5022,6 +5146,7 @@ static const struct dt_object_operations osd_obj_ops = {
 	.do_xattr_list		= osd_xattr_list,
 	.do_object_sync		= osd_object_sync,
 	.do_invalidate		= osd_invalidate,
+	.do_check_stale		= osd_check_stale,
 };
 
 static const struct dt_object_operations osd_obj_otable_it_ops = {
@@ -5478,8 +5603,7 @@ static int __osd_ea_add_rec(struct osd_thread_info *info,
 	LASSERT(pobj->oo_inode);
 
 	ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
-	if (unlikely(pobj->oo_inode ==
-		     osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
+	if (unlikely(osd_object_is_root(pobj)))
 		ldp->edp_magic = 0;
 	else
 		osd_get_ldiskfs_dirent_param(ldp, fid);
@@ -5640,11 +5764,9 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
 
 static int
 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
-		      struct osd_idmap_cache *oic)
+		      const struct lu_fid *fid, struct osd_inode_id *id)
 {
 	struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
-	struct lu_fid *fid = &oic->oic_fid;
-	struct osd_inode_id *id = &oic->oic_lid;
 	struct inode *inode = NULL;
 	int once = 0;
 	bool insert;
@@ -5707,7 +5829,7 @@ trigger:
 			}
 		}
 
-		rc = osd_oii_insert(dev, oic, insert);
+		rc = osd_oii_insert(dev, fid, id, insert);
 		/*
 		 * There is race condition between osd_oi_lookup and OI scrub.
 		 * The OI scrub finished just after osd_oi_lookup() failure.
@@ -5720,7 +5842,7 @@ trigger:
 		if (!S_ISDIR(inode->i_mode))
 			rc = 0;
 		else
-			rc = osd_check_lmv(oti, dev, inode, oic);
+			rc = osd_check_lmv(oti, dev, inode);
 
 		GOTO(out, rc);
 	}
@@ -5729,8 +5851,7 @@ trigger:
 		rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
 				     SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 		CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
-		       "%s: trigger partial OI scrub for RPC inconsistency "
-		       "checking FID "DFID": rc = %d\n",
+		       "%s: trigger partial OI scrub for RPC inconsistency checking FID "DFID": rc = %d\n",
 		       osd_dev2name(dev), PFID(fid), rc);
 		if (rc == 0 || rc == -EALREADY)
 			goto again;
@@ -5739,18 +5860,17 @@ trigger:
 	GOTO(out, rc);
 
 out:
-	if (inode)
-		iput(inode);
+	iput(inode);
 
 	RETURN(rc);
 }
 
 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
 			       struct osd_device *dev,
-			       struct osd_idmap_cache *oic,
 			       struct lu_fid *fid, __u32 ino)
 {
 	struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
+	struct osd_idmap_cache *oic = &oti->oti_cache;
 	struct inode *inode;
 	int rc;
 
@@ -5941,13 +6061,12 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
 	if (!IS_ERR(bh)) {
 		struct osd_thread_info *oti = osd_oti_get(env);
 		struct osd_inode_id *id = &oti->oti_id;
-		struct osd_idmap_cache *oic = &oti->oti_cache;
 		struct osd_device *dev = osd_obj2dev(obj);
 
 		ino = le32_to_cpu(de->inode);
 		if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
 			brelse(bh);
-			rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
+			rc = osd_fail_fid_lookup(oti, dev, fid, ino);
 			GOTO(out, rc);
 		}
 
@@ -5975,19 +6094,24 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
 			osd_id_gen(id, ino, OSD_OII_NOGEN);
 		}
 
-		if (rc != 0 || osd_remote_fid(env, dev, fid)) {
-			fid_zero(&oic->oic_fid);
-
+		if (rc != 0 || osd_remote_fid(env, dev, fid))
 			GOTO(out, rc);
-		}
 
-		osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
-		rc = osd_consistency_check(oti, dev, oic);
-		if (rc == -ENOENT)
-			fid_zero(&oic->oic_fid);
-		else
+		rc = osd_consistency_check(oti, dev, fid, id);
+		if (rc != -ENOENT) {
 			/* Other error should not affect lookup result. */
 			rc = 0;
+
+			/* Normal file mapping should be added into OI cache
+			 * after FID in LMA check, but for local files like
+			 * hsm_actions, their FIDs are not stored in OI files,
+			 * see osd_initial_OI_scrub(), and here is the only
+			 * place to load mapping into OI cache.
+			 */
+			if (!fid_is_namespace_visible(fid))
+				osd_add_oi_cache(osd_oti_get(env),
+						 osd_obj2dev(obj), id, fid);
+		}
 	} else {
 		rc = PTR_ERR(bh);
 	}
@@ -6690,18 +6814,13 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
 	filp->f_cred = current_cred();
 	rc = osd_security_file_alloc(filp);
 	if (rc)
-		RETURN(rc);
+		GOTO(unlock, rc);
 
 	filp->f_flags |= O_NOATIME;
 	filp->f_mode |= FMODE_NONOTIFY;
 	rc = iterate_dir(filp, &buf.ctx);
 	if (rc)
-		RETURN(rc);
-
-	if (hlock != NULL)
-		ldiskfs_htree_unlock(hlock);
-	else
-		up_read(&obj->oo_ext_idx_sem);
+		GOTO(unlock, rc);
 
 	if (it->oie_rd_dirent == 0) {
 		/*
@@ -6715,6 +6834,11 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
 		it->oie_dirent = it->oie_buf;
 		it->oie_it_dirent = 1;
 	}
+unlock:
+	if (hlock != NULL)
+		ldiskfs_htree_unlock(hlock);
+	else
+		up_read(&obj->oo_ext_idx_sem);
 
 	RETURN(rc);
 }
@@ -6863,11 +6987,10 @@ osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
 	 * That means we lose it!
 	 */
 	if (rc != 0)
-		CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
-		       "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
-		       osd_ino2name(inode),
-		       dir->i_ino, dir->i_generation, namelen,
-		       dentry->d_name.name, PFID(fid), rc);
+		CDEBUG(D_LFSCK,
+		       "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
+		       osd_ino2name(inode), dir->i_ino, dir->i_generation,
+		       namelen, dentry->d_name.name, PFID(fid), rc);
 
 	RETURN(rc);
 }
@@ -7597,11 +7720,11 @@ static int osd_mount(const struct lu_env *env,
 	const char *name = lustre_cfg_string(cfg, 0);
 	const char *dev = lustre_cfg_string(cfg, 1);
 	const char *opts;
-	unsigned long page, s_flags, lmd_flags = 0;
+	unsigned long page, s_flags = 0, lmd_flags = 0;
 	struct page *__page;
 	struct file_system_type *type;
 	char *options = NULL;
-	char *str;
+	const char *str;
 	struct osd_thread_info *info = osd_oti_get(env);
 	struct lu_fid *fid = &info->oti_fid;
 	struct inode *inode;
@@ -7616,11 +7739,9 @@ static int osd_mount(const struct lu_env *env,
 		RETURN(-E2BIG);
 	strcpy(o->od_mntdev, dev);
 
-	str = lustre_cfg_string(cfg, 2);
-	s_flags = simple_strtoul(str, NULL, 0);
-	str = strstr(str, ":");
-	if (str)
-		lmd_flags = simple_strtoul(str + 1, NULL, 0);
+	str = lustre_cfg_buf(cfg, 2);
+	sscanf(str, "%lu:%lu", &s_flags, &lmd_flags);
+
 	opts = lustre_cfg_string(cfg, 3);
 #ifdef __BIG_ENDIAN
 	if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
@@ -7670,6 +7791,7 @@ static int osd_mount(const struct lu_env *env,
 			"force_over_256tb",
 			"force_over_512tb",
 			"force_over_1024tb",
+			"resetoi",
 			NULL
 		};
 		strncat(options, opts, PAGE_SIZE);
@@ -7725,7 +7847,7 @@ static int osd_mount(const struct lu_env *env,
 	}
 
 	if (lmd_flags & LMD_FLG_DEV_RDONLY) {
-		LCONSOLE_WARN("%s: not support dev_rdonly on this device",
+		LCONSOLE_WARN("%s: not support dev_rdonly on this device\n",
 			      name);
 
 		GOTO(out_mnt, rc = -EOPNOTSUPP);
@@ -7787,6 +7909,8 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
 	osd_procfs_fini(o);
 	if (o->od_oi_table != NULL)
 		osd_oi_fini(osd_oti_get(env), o);
+	if (o->od_extent_bytes_percpu)
+		free_percpu(o->od_extent_bytes_percpu);
 	osd_obj_map_fini(o);
 	osd_umount(env, o);
 
@@ -7799,8 +7923,10 @@ static int osd_device_init0(const struct lu_env *env,
 {
 	struct lu_device *l = osd2lu_dev(o);
 	struct osd_thread_info *info;
-	int rc;
 	int cplen = 0;
+	char *opts = NULL;
+	bool restored = false;
+	int rc;
 
 	/* if the module was re-loaded, env can loose its keys */
 	rc = lu_env_refill((struct lu_env *)env);
@@ -7828,6 +7954,8 @@ static int osd_device_init0(const struct lu_env *env,
 	o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
 	o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
 	o->od_auto_scrub_interval = AS_DEFAULT;
+	/* default fallocate to unwritten extents: LU-14326/LU-14333 */
+	o->od_fallocate_zero_blocks = 0;
 
 	cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
 			sizeof(o->od_svname));
@@ -7865,10 +7993,14 @@ static int osd_device_init0(const struct lu_env *env,
 	if (rc != 0)
 		GOTO(out_site, rc);
 
+	opts = lustre_cfg_string(cfg, 3);
+	if (opts && strstr(opts, "resetoi"))
+		restored = true;
+
 	INIT_LIST_HEAD(&o->od_ios_list);
 	/* setup scrub, including OI files initialization */
 	o->od_in_init = 1;
-	rc = osd_scrub_setup(env, o);
+	rc = osd_scrub_setup(env, o, restored);
 	o->od_in_init = 0;
 	if (rc < 0)
 		GOTO(out_site, rc);
@@ -7910,6 +8042,12 @@ static int osd_device_init0(const struct lu_env *env,
 		GOTO(out_procfs, rc);
 	}
 
+	o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
+	if (!o->od_extent_bytes_percpu) {
+		rc = -ENOMEM;
+		GOTO(out_procfs, rc);
+	}
+
 	RETURN(0);
 
 out_procfs:
@@ -8257,7 +8395,7 @@ static int __init osd_init(void)
 		(void *)kallsyms_lookup_name("security_file_alloc");
 #endif
 
-	rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
+	rc = class_register_type(&osd_obd_device_ops, NULL, true,
 				 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
 	if (rc) {
 		lu_kmem_fini(ldiskfs_caches);